Sei sulla pagina 1di 14

Applied Thermal Engineering 23 (2003) 21692182

www.elsevier.com/locate/apthermeng

Thermoeconomic optimization of combined cycle gas turbine power plants using genetic algorithms
Manuel Vald es *, Ma Dolores Dur an, Antonio Rovira
tica y Fluidomeca nica, Universidad Polite cnica de Madrid, Departamento de Ingenier a Energe  Gutie rrez Abascal 2, 28006 Madrid, Spain E.T.S. Ingenieros Industriales, Jose Received 20 March 2003; accepted 13 June 2003

Abstract This paper shows a possible way to achieve a thermoeconomic optimization of combined cycle gas turbine (CCGT) power plants. The optimization has been done using a genetic algorithm, which has been tuned applying it to a single pressure CCGT power plant. Once tuned, the optimization algorithm has been used to evaluate more complex plants, with two and three pressure levels in the heat recovery steam generator (HRSG). The variables considered for the optimization were the thermodynamic parameters that establish the conguration of the HRSG. Two dierent objective functions are proposed: one minimizes the cost of production per unit of output and the other maximizes the annual cash ow. The results obtained with both functions are compared in order to nd the better optimization strategy. The results show that it is possible to nd an optimum for every design parameter. This optimum depends on the selected optimization strategy. 2003 Elsevier Ltd. All rights reserved.
Keywords: Thermoeconomics; Combined cycle gas turbine; Heat recovery steam generator; Genetic algorithms

1. Introduction If the sole objective of a CCGT design was to maximize the thermodynamic eciency, its total cost would be very high. This is why many authors [14] have made thermoeconomic studies in order to get a compromise between plant eciency and generation costs.

Corresponding author. Fax: +34-1-3363006. E-mail addresses: mvaldes@etsii.upm.es (M. Vald es), mduran@eneru.etsii.upm.es (Ma .D. Dur an), rovira@motores.etsii.upm.es (A. Rovira). 1359-4311/$ - see front matter 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S1359-4311(03)00203-5

2170

M. Vald es et al. / Applied Thermal Engineering 23 (2003) 21692182

Nomenclature A HRSG area (m2 ) B simplied cash ow (/year) amortization cost per year (/year). Ca fuel cost per kW h (/kW h) Cf total cost of the fuel (/year) CTf gas turbine xed cost () CGT steam turbine xed cost () CST CHRSG HRSG xed cost () CkW h generation cost (/kW h) operation and maintenance cost per year (/year) Com total cost per year (/year) CTot UA HRSG area increment xed cost () Cf-HRSG h plant working hours per annual plant operation period (h/year) total income (/year) ITot k cost of a single HRSG section per product UA (K/kW) air mass ow (kg/s) ma gas turbine exhaust mass ow (kg/s) mg fuel mass ow (kg/s) mf N CCGT economic life (years) Pen penalization function PP pinch point (K) AP approach point (K) S selling price of kW h (/kW h) U overall heat transfer coecient (W m2 K1 ) W gross power (kW) W mean annual output of the plant (kW) X moisture (%) Dp pressure loss (mbar) DT gas to steam temperature dierence at superheater (K) g thermal eciency Subscripts HP high pressure IP intermediate pressure LP low pressure amb ambient conditions j number of design variables inl inlet conditions exh exhaust conditions ec economizer ev evaporator

M. Vald es et al. / Applied Thermal Engineering 23 (2003) 21692182

2171

sh superheater section section of the HRSG

For example, Attala et al. [1] have optimized a dual pressure level CCGT. They have worked with a simulation program with three modules: the rst one simulates the cycle, the second one evaluates the thermodynamic and thermoeconomic parameters and the last one is the optimization model. Likewise, the target of this paper is to present a CCGT thermoeconomic optimization model. This analysis aims to obtain a good design of the heat recovery steam generator (HRSG). It has been decided so because the HRSG is one of the most important components of the plant and its n design aects to a large extent the eciency and the cost of the whole plant, as Vald es and Rapu [5] have shown. In the present work a particular gas turbine is chosen and its outlets (i.e. mass ow mg and the temperature of the available ue gas Texh-GT ) are used as input data for the HRSG design, which in turn permits the calculation of the steam turbine and therefore of the whole power plant. Then it is possible to quantify the eect of the changes in the design parameters on the whole cycle eciency or on the individual components performances. A rough approach would be to carry out parametric studies, running repeatedly the simulation code while introducing suitable changes in a given parameter of the equations involved. The use of an optimization technique is a more powerful approach to assess in a direct way the eect of design modications. An optimization procedure using inuence coecients derived from the solution of the system equations through the NewtonRaphson method was already tested [5]. Nevertheless, while this is a powerful iteration technique, it should be used carefully because if the rst iteration is too far away from the solution, the iterative process may not converge. Likewise, if any of the iterations ends up in a solution far o the real one, the method might not continue or even nish in a nonphysical solution of the system of equations. There are other techniques like gradient methods, which need explicit objective functions, or complex search methods, which need a clearly dened searching space and require much calculation time. Instead, it was decided to use a genetic algorithm optimization method (GA), described in Appendix A. This is a generic method that uses the penalty function approach and has the advantage that it can be easily applied to almost any optimization problem. Previous applications of this method by other authors [1,6,7] have yielded successful results.

2. Thermoeconomic model 2.1. Description of the combined cycles used There are many dierent design alternatives for CCGT power plants due to the large quantity of design parameters that can be taken into consideration; for example, the number of pressure levels, the distribution of economizers, evaporators and superheaters into the HRSG and the use of reheaters or preheaters.

2172

M. Vald es et al. / Applied Thermal Engineering 23 (2003) 21692182

In order to cover a wide variety of alternatives, the thermoeconomic optimization model has been applied to the following congurations: 1. single pressure CCGT power plant, 2. dual pressure CCGT power plant with and without reheating, 3. triple pressure CCGT power plant with reheating. Single pressure CCGT power plant. Since the single pressure CCGT power plant is the simplest conguration of all, it has been used to t the genetic algorithm optimization method. The energytemperature diagram and a schematic of the HRSG of this cycle is shown in Fig. 1a. Dual and triple pressure CCGT power plants. Dual and triple pressure CCGT power plants are currently being installed worldwide. Fig. 1bd show one of the possible congurations of the dual pressure (with and without reheating) and of the triple pressure CCGT power plants. In these gures the pinch points, approach points and the gas to steam temperature dierence at the superheaters are also dened. 2.2. Thermoeconomic objective functions Two dierent optimization criteria have been considered: on the one hand the maximization of the annual cash ow and on the other hand the minimization of the generation cost of the plant. The annual cash ow function is: 1 f xj B ITot CTot The generation cost function is the mean annual energy output divided by the total cost per year. W h 1=CkW h 2 CTot The goal of the optimization algorithm is to maximize the functions f xj . For identity (2), this implies the minimization of the generation cost CkW h . The functions f xj greatly depend on the eciency of the cycle, but in dierent ways, and the results of the optimizations should be dierent depending on the function chosen. For example, as it is shown in Section 5, the optimized total heat transfer area of the HRSG is lower for the minimum cost than for the maximum cash ow solution. The total income in Eq. (1) is dened by: f xj ITot S W h 3 The total cost per year includes the fuel cost, the amortization cost and the operation and maintenance cost: 4 CTot CTf Ca Com where CTf is the total fuel cost:   W h 5 CTf Cf g Ca is the amortization cost that accounts for the total xed cost divided by the plant economic life time:

M. Vald es et al. / Applied Thermal Engineering 23 (2003) 21692182

2173

Fig. 1. Energy temperature diagram and schematic of the congurations of the power plants considered. (a) Single pressure CCGT (1P); (b) dual pressure CCGT (2P); (c) dual pressure CCGT with reheating (2PR); (d) triple pressure CCGT with reheating (3PR).
UA CGT CST CHRSG Cf-HRSG N

Ca

Ca includes two terms. The rst one is a xed cost that is characteristic of the CCGT components (gas turbine, steam turbine and HRSG). The second one is the cost associated with a possible

2174

M. Vald es et al. / Applied Thermal Engineering 23 (2003) 21692182

variation of the heat transfer area due to the changes that take place in the HRSG design parameters during the optimization process. It is also a xed cost which is assumed to be proportional to the product of the overall heat transfer coecient U and the HRSG area A:
UA ksection UAsection Cf-HRSG ;section

Considering three dierent types of sections (economizers, evaporators and superheaters) Eq. (7) becomes: X X X UA kec UAec kev UAev ksh UAsh 8 Cf-HRSG
ec ev sh

Com is the operation and maintenance cost. As pointed out by Naugheten [8] it can be assumed to be the 10% of the total plant cost: Com 0:10 CTot 9

3. General layout of the optimization model The solution of every optimization problem involves the selection of the design variables, the objective functions and the constraints of the problem. The design variables are the thermodynamic parameters of the HRSG that can take independent values during the design process: the dierent pressure levels and the temperatures (PP, AP and DT ) that are shown in Fig. 1. The design variables may change within the optimization interval according to previously established limits. Table 1 shows the limits selected for each conguration. Most of these limits were determined according to the authors experience, while some values were taken from Horlock [9].
Table 1 Limit values of the design variables Single pressure Limits PHP (bar) PPHP (K) APHP (K) DTHP (K) PIP (bar) PPIP (K) APIP (K) DTIP (K) PLP (bar) PPLP (K) APLP (K) DTLP (K) Low 20 3 3 25 High 60 8 8 80 Dual pressure Low 50 5 3 40 3 3 3 20 High 100 17 20 80 12 15 15 120 Dual pressure + reheat Low 40 5 4 10 5 5 4 15 High 70 17 20 50 20 20 30 80 Triple pressure + reheat Low 100 3 3 40 25 3 3 40 4 3 3 High 160 15 15 70 50 15 30 80 10 10 10

M. Vald es et al. / Applied Thermal Engineering 23 (2003) 21692182

2175

Once a particular set of thermodynamic design parameters has been selected, the heat transfer area of the dierent HRSG sections and the performance of the steam cycle may be calculated n [5]. following Vald es and Rapu In a general optimization problem, along with the objective function, the constraints of the problem must be dened clearly. When using a genetic algorithms method, the tness function includes the objective function and a number of penalization functions which depend on the physical constraints of some dependent variables. Without the penalizations it would be nearly impossible to nd an optimum. In order to dene these penalizations the following considerations have been done: The exhaust gas temperature of the HRSG Texh-HRSG should not be below 100 C. The moisture content X in the last stage of the steam turbine should not be over 16%. With these conditions the penalization functions are dened as:  AbsTexh-HRSG 100 if Texh-HRSG 6 100 Pen Texh-HRSG 0 if Texh-HRSG > 100  Abs0:16 X if X P 0:16 Pen X 0 if X < 0:16 Then the tness function is written in the following way: ffitness xj f xj P1 Pen Texh-HRSG P2 Pen X 12 where xj design variables; f xj objective function dened by identities (1) or (2); P1 and P2 constants. Their values should be high enough to remove the individuals 1 that give a value outside the established interval; j 1; . . . ; number of design variables.

10 11

4. Optimization program 4.1. Description of the program Prior to the optimization process, it is necessary to model the system and to select the optimization variables. For that reason a Visual Basic program was built. It includes the following four modules: 1. Gas turbine simulation: This module simulates the gas turbine cycle in order to calculate its outlets at part and full load. The design of the gas turbine is not part of the optimization model. 2. Conguration selection: The geometry of the HRSG is dened in this module. The objective functions and the restrictions of the design variables are also selected in this step.

Individuals are the random design variables generated by the genetic algorithm (see Appendix A).

2176

M. Vald es et al. / Applied Thermal Engineering 23 (2003) 21692182

3. Design and simulation of the CCGT: It is used to calculate the operational variables (mass ow, eciency, moisture content, etc.). This module includes all the equations that govern the performance of the dierent components of the system. 4. Optimization of the CCGT power plant: Once the assignment of the objective function is done, the GA optimization tool (described in Appendix A) is employed to optimize the cycle. The tness (health) of each individual is found using the results of the third module described above. 4.2. Parameters of the genetic algorithm As described in Appendix A, the genetic algorithm (GA) needs a number of parameters: Population, number of generations, mutation percentage and crossover percentage. Following Qi-Wen et al. [10] and Samir [11], a correct selection of the GA parameters is very important for the convergence of the algorithm although this selection depends on the particular conditions of each problem. Qi-Wen considers that it is necessary to test the algorithm varying these parameters in order to know which values make the algorithm more stable. As stated before, the single pressure CCGT conguration was used to determine the right values of the GA parameters because it is the simplest conguration of CCGT power plants. This GA parameters were subsequently used in the optimization of the other power plants.

5. Results 5.1. Results of the thermoeconomic optimization The thermoeconomic optimization of the four dierent congurations tested was done according to the specications mentioned in Section 2.1. Table 2 shows the design point conditions of the gas turbine as well as the results of the gas cycle simulation. These results were used as input data for the optimization process. Table 3 shows the optimization results obtained applying optimum cost and optimum cash ow solutions. In both cases the single pressure CCGT power plant results (generation cost and cash ow) are used as the 100% point of reference against which the results of the other cycle congurations are compared. It can be noticed that the optimization results applying the cash ow function lead to a slightly higher value of the generation costs. Likewise, the optimum cost solution leads to a lower cash ow. For some variables 2 the optimum value matches the maximum or minimum of the established limits of the design variables. This means that the optimization algorithm yields the limiting operative value selected for that variable. For the three pressure level CCGT power plant the optimum found by the algorithm has lower eciency than for the dual pressure case. This is because the cost of the HRSG for the three
2

The variables whose optimum value coincides with the constraint established are marked with an asterisk.

M. Vald es et al. / Applied Thermal Engineering 23 (2003) 21692182 Table 2 GT data at the design point Parameter Compression ratio Compressor isentropic eciency Turbine isentropic eciency Tamb (K) Pamb (bar) Combustion chamber eciency Combustion chamber pressure loss (%) Tinl (K) ma (kg/s) WGT (MW) gGT mf (kg/s) Texh-GT (K) Input data 20 0.85 0.91 288.15 1.013 0.95 4 1430 300 105.3 0.382 5.69 776.3 Performances

2177

Table 3 Results of the thermoeconomic optimization Single pressure Optimum cost B (%) CkW h (%) g PHP (bar) PPHP (K) APHP (K) DTHP (K) PIP (bar) PPIP (K) APIP (K) DTIP (K) PLP (bar) PPLP (K) APLP (K) DTLP (K) 99.90 100 50.58 37.81 7.40 3.03* 67.23 Optimum cash ow 100 100.3 50.65 37.26 6.15 3.05* 63.29 Dual pressure Optimum cost 147.70 93.35 54.35 97.34 14.41 5.91 45.86 4.26 14.69 3.84 105.63 Optimum cash ow 148.04 93.41 54.54 94.20 10.88 3.36* 45.85 4.25 12.61 3.54 102.46 Dual pressure + reheat Optimum cost 138.70 93.74 53.46 61.80 15.20 17.67 40.15 10.56 5.95 4.46 23.59 Optimum cash ow 148.25 94.15 54.20 61.06 14.85 6.80 15.45 5.17* 7.43 8.87 15.06* Triple pressure + reheat Optimum cost 98.54 100.3 51.05 132.05 6.46 6.36 50.28 30.05 10.51 6.76 55.01 4.01* 6.82 4.72 Optimum cash ow 99.32 100.94 51.05 138.66 6.94 6.90 51.58 30.09 10.90 4.73 50.32 4.02* 6.26 5.72

* Optimum value matches the limit of the variable.

pressure level power plant rapidly increases if a higher value of eciency is desired. Therefore the optimization algorithm nds a better solution, from an economic point of view, with a lower eciency value. Also, it can be observed that, for the particular gas turbine used in this work, the optimum design for the three pressure level conguration has a higher generation cost and a lower cash ow than the dual pressure congurations. This does not mean that, in general, a triple pressure CCGT power plant yields poorer results than the others, but rather that, in this particular case, it is more interesting to use the dual pressure CCGT power plant.

2178

M. Vald es et al. / Applied Thermal Engineering 23 (2003) 21692182

In general it was noticed that, for the entry conditions established in this work, the congurations which give better results are the dual pressure CCGT power plant with and without reheating. For other entry conditions the results might be dierent. Finally, it was observed that if the cash ow is selected as the objective function then the dimensions, and consequently the cost, of the HRSG optimum design will be higher. This is because, in this case, the increment of eciency overcomes the cost increment as higher incomes are obtained. 5.2. Parametric study Finally, a parametric study was done in order to check if true optimum values were reached and the algorithm did not converge into a local singularity. The study was done for the single pressure conguration. Figs. 2 and 3 show the variation around the optimum value for the cost function and the cash ow function respectively, while Fig. 4 shows the variation of the eciency in the vicinity of the optimum cost point. Unlike the cost and the cash ow, the eciency does not have an optimum value. Besides, it may be seen that, for this case, the variable that has more eect on the cost and the cash ow is the drum pressure, while the pinch point is the variable that has more eect on the eciency. At rst glance it would seem that any variation on the considered design parameters has little inuence on the system performance. This is primarily due to the small variation of the eciency when the design parameters are near their optima (where the objective function curves are quite at). Anyway, for some parameters like the pinch point, a high rate of growth in percentage terms can be reached with an increase in the absolute value of only a few degrees centigrade. On the other hand, it must be pointed out that small percentage variations in the cost or cash ow functions can inuence to a large extent the economic results of the plant. For example, for the particular case of the two pressure cycle with reheat studied in this work, a 1% variation in cash ow (starting from the optimum) is equivalent to 120,000 /year.
100.07 100.06

Relative cost (%)

100.05 100.04 100.03 100.02 100.01 100 99.99 -30 -20 -10 0 10 20 30

Pressure Pinch Point

Temp. Difference Approach Point

Fig. 2. Variation of the generation cost around the optimum design point.

M. Vald es et al. / Applied Thermal Engineering 23 (2003) 21692182


100.05 100

2179

Relative cash flow (%)

99.95 99.9 99.85 99.8 99.75 99.7 99.65 -30 -20 -10 0 10 20 30

Pressure Pinch Point

Temp. Difference Approach Point

Fig. 3. Variation of the cash ow around the optimum design point.

0.507 0.5068 0.5066 0.5064

Efficiency

0.5062 0.506 0.5058 0.5056 0.5054 0.5052 0.505 0.5048 -30 -20 -10 0 10 20 30

Pressure Pinch Point

Temp. Difference Approach Point

Fig. 4. Variation of the eciency around the optimum cost design point.

5.3. Applicability of the algorithm In order to check whether the optimization algorithm works well, the evolution of the generations during the selection process is presented in Fig. 5. The z axis is the generation number while x and y are the values of the pinch point and the drum pressure respectively 3 for each individual
The pinch point and the drum pressure were selected because they were thought to have the greatest inuence on the CCGT performance.
3

2180

M. Vald es et al. / Applied Thermal Engineering 23 (2003) 21692182


P 40 30 50 60 4 PP 6

8 150

100 Gen

50

Healthiest Medium healthy Least healthy

Fig. 5. Evolution of the genetic algorithm.

of a single pressure cycle. The least healthy individuals disappear as the optimization process takes place. It was also observed that the algorithm converges rapidly. Fig. 6 shows the evolution of the maximum cash ow value corresponding to each generation (the maximum is the 100% reference value). It can be noticed that the maximum value found in the rst generation is similar to the one found in the last generation. This means that the algorithm selected a healthy individual from the beginning. Fig. 7 shows the maximum and mean value corresponding to each generation taking the average cash ow as the objective function. The mean value varies despite the algorithm has already found the optimum. These variations of the mean value correspond to the appearance of a less healthy individual inside the population, caused by the mutation operator. However, it can be noticed that in general the dierence between the mean and the maximum value decreases as generations go by.
100.01

Relative cash flow (%)

100 99.99 99.98 99.97 99.96 99.95 99.94 99.93 0 50 100 Generations 150

Maximum relative cash flow

Fig. 6. Evolution of the best individual of each generation.

M. Vald es et al. / Applied Thermal Engineering 23 (2003) 21692182


Relative cash flow (%)

2181

99.9 99.7 99.5 99.3 99.1 98.9 98.7 98.5 0 50 100 Generations Mean 150

Maximum

Fig. 7. Evolution of the maximum and mean values of each generation.

6. Conclusions The genetic algorithm optimization tool can be successfully applied to the optimization of CCGT power plants. Indeed it could be possible to apply it even in a multiobjective optimization problem. With this optimization method it is possible to identify the set of design variables that is inuencing the objective function to a greater extent. Particularly, for the one pressure level CCGT conguration, the pinch point and the drum pressure have higher eects on the cost than the approach point and the gas to steam temperature dierence at the superheater. The optimization using two dierent criteria is better suited for the design of this kind of systems than a single optimization criterion because this approach shows whether it is worth reducing the benet in order to have lower generation costs or not. It has been pointed out that the sum of the UA of the dierent HRSG sections tends to be greater if the intention is to optimize the cash ow, particularly for the single pressure cycle. In this case the cost increase is overcome with an increase in the production and the incomes owing to the higher heat exchange area. As it was expected, the single pressure CCGT power plant is the worst in terms of costs and cash ow and has the lowest eciency value because the energy is poorly used with this conguration. The optimization algorithm has selected an optimum design for the triple pressure CCGT power plant with lower eciency than the corresponding dual pressure. This means that a search for a higher eciency in this conguration would increase the xed costs too much for the particular conditions selected in this work. Acknowledgement The authors wish to acknowledge the nancial support of ENDESA to this work. Appendix A. Description of the genetic algorithms tool One of the objectives of this work is to set up a methodology that facilitates the design and optimization of the CCGT. The algorithm applied here has been already used by other authors [1].

2182

M. Vald es et al. / Applied Thermal Engineering 23 (2003) 21692182

In general, genetic algorithms work in the way described below (a better description of these algorithms may be found in Goldberg [12,13] and Bentley [14]): 1. A population of a certain number of individuals is randomly generated. The individuals are identied by the values of the design variables. 2. All the individuals are evaluated with the tness function and they are classied according to this value. 3. The healthiest individuals are selected as the parents of the following generation. Genetic operators (mutation and crossover) are applied to these selected individuals and a new generation is obtained. Each generation has the same population size. 4. The new generation is evaluated again with the tness function. The hypothesis underlying this method is that the new generation is formed by healthier individuals than the previous one. 5. Finally, the process is repeated until a previously established number of generations is reached.

References
[1] L. Attala, B. Facchini, G. Ferrara, Thermoeconomic optimization method as design tool in gas-steam combined plant realization, Energy Conversion and Management 42 (2001) 21632172. [2] P.J. Dechamps, Incremental cost optimization of heat recovery steam generators, ASME COGEN-TURBO 95. [3] G. Tsatsaronis, Thermoeconomic analysis and optimization of energy systems, Progress in Energy and Combustion Science 19 (1993) 227257. [4] A. Franco, A. Russo, Combined cycle plant eciency increase based on the optimization of the heat recovery steam generator operating parameters, International Journal of Thermal Sciences 41 (2002) 843859. n, Optimization of a heat recovery steam generators for combined cycle gas turbine power [5] M. Vald es, J. Rapu plants, Applied Thermal Engineering 21 (2001) 11491159. [6] A. Toolo, A. Lazzareto, Evolutionary algorithms for multi-objective energetic and economic optimization in thermal system design, Energy 27 (2002) 549567. [7] G. Fabbri, Optimization of heat transfer through nned dissipators cooled by laminar ow, International Journal of Heat and Fluid Flow 19 (6) (1998) 644654. [8] B. Naugheten, Economic assessment of combined cycle gas turbines in Australia, some eects of microeconomic reform and technological change, Energy Policy 31 (3) (2003) 225245. [9] J.H. Horlock, Combined Power Plants, rst ed., Pergamon Press, Great Britain, 1992. [10] Y. Qi-Wen, J. Jing-Ping, Guo Chen, How to select optimum control parameters for genetic algorithms, Proceedings of the ISIE-IEEE, 2001, pp. 3741. [11] W.M. Samir, Niching methods for genetic algorithms, Illigal Report 95001, May 1995. [12] D. Goldberg, M.P. Santani, Engineering optimization via genetic algorithm, in: Proceedings of the Ninth Conference on Electronic Computation, ASCE, New York, 1986, pp. 471482. [13] D. Goldberg, Genetic Algorithms in Search Optimization and Machine Learning, Addison-Wesley, Michigan, 1996. [14] P. Bentley, An Introduction to Evolutionary Design by Computers; Evolutionary Design by Computers, 1999, pp. 171.

Potrebbero piacerti anche